Autonomous Passive Localization Algorithm for Shared Bikes

Opportune and accurate position information of the ubiquitous shared bikes can help users rapidly find the available bicycle around them. In order to remove the professional locating device of bicycles and reduce industrial costs, a passive positioning method for shared bikes named PSA is proposed in this paper with the support of activity recognition based on deep learning. First, an action recognition module based on LSTM algorithm is constructed through training three-axis accelerator data collected by commercial mobile phone to distinguish user's riding state from other actions like walking or running; then, the GPS module embedded in the user's mobile phone will record the location when the state changes from riding to non-riding, which is regarded as the bike's return location. Furthermore, considering the variety of state changing during riding procedure, a duration threshold mechanism is added for the sake of fault tolerance processing. Several simulations and test-bed experiments were conducted to verify the feasibility and practicality of the proposed method, where the accuracy of activity recognition in practical data set can reach more than 95% and the detection time in public data set is less than 0.2s. In addition, this method reduces phone operations and enhances the convenience of the experience.